This paper proposes a statistical method to detection of Topological Charges (TCs) based on calculating the similarity factor (SF) between the instantaneous crosstalk distribution and the standard crosstalk distributions. The values of SF are time-varying, and we set a threshold to recognize TCs. Then, the probability density function (PDF) of the SF is obtained to estimate error rate of TC detection under weak, moderate and strong turbulence condition. At last, we investigate the influence of the length of training sequence on the standard crosstalk distribution for the practical applications.